Artificial Intelligence In Services Addressing Violence Perpetration Among Adolescents and Young Adults: A Scoping Review

Background and Purpose: Adolescence and young adulthood (ages 12–25) represent a critical developmental window where sensitivity to environmental influences can shape long-term behavioral trajectories. Aggression and violence during this period remain significant global public health challenges. While artificial intelligence (AI) has demonstrated transformative potential in enhancing administrative workflows, screening, and treatment planning within general healthcare, its integration into services specifically targeting youth violence perpetration remains fragmented. This scoping review systematically maps the current landscape of AI applications in youth violence services to identify targeted violence outcomes, summarize employed AI techniques, and highlight critical methodological and ethical gaps.

Methods: Following PRISMA-ScR guidelines, a comprehensive search was conducted across six electronic databases (Web of Science, MEDLINE, PubMed, EMBASE, APA PsycINFO, and Social Work Abstracts) for peer-reviewed studies published between 1991 and 2025. Studies were eligible if they focused on AI applications for prediction, detection, diagnosis, treatment/intervention, or prognosis services for violence perpetrators aged 12–25. Four independent reviewers conducted screening and data extraction. Findings were synthesized narratively and supported by descriptive analyses.

Results: Sixteen studies met the inclusion criteria. Most were conducted in developed countries (n=11), primarily the United States (n=6), with only two studies from developing country (China). Participant samples varied widely; nine studies involved thousands of individuals, while three had fewer than 100. AI applications were heavily skewed toward detection and prediction services (n=14), while only two studies addressed treatment-oriented interventions and none focused on administrative planning or prognosis. Common violence types addressed included cyberbullying (n=5), crime or intentional injuries (n=4), and school violence (n=2). However, more private forms of harm, such as intimate partner violence (IPV), were rarely investigated. Technologically, machine learning was the dominant approach (n=11), followed by deep learning and neural networks (n=6). Data sources ranged from social media text and audio recordings to clinical records and sensor data. While several studies reported that AI outperformed traditional statistical models or clinical judgment, significant reporting gaps remained. Notably, 11 studies omitted socioeconomic status, 10 omitted race/ethnicity, 10 did not specify participant gender, and four did not report legal or privacy protections.

Conclusions and Implications: This review demonstrates that while AI use in youth violence services is growing, it remains largely restricted to predictive functions for digitally observable behaviors. For social work practice and policy, these findings highlight a critical need to pivot research toward AI-supported interventions and therapeutic strategies. Furthermore, the lack of demographic diversity and ethical transparency in current datasets poses a risk of algorithmic bias. Future efforts must prioritize inclusive data infrastructures, robust ethical safeguards, and multidisciplinary collaborations between data scientists and social service providers to ensure that AI tools effectively and equitably support the well-being of diverse young populations.